cecli-dev 0.97.2__tar.gz → 0.97.4__tar.gz

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cecli-dev
-Version: 0.97.2
+Version: 0.97.4
 Summary: we also can't pronounce cecli
 Project-URL: Homepage, https://github.com/dwash96/cecli
 Project-URL: Documentation, https://github.com/dwash96/cecli
@@ -84,12 +84,12 @@ Dynamic: license-file
 
 ## Why `cecli`?
 
-`cecli` (probably pronounced like "Cecily", aka `aider-ce`) is a community-driven fork of the [Aider](https://cecli.dev/) AI pair programming tool.
-Aider is a fantastic piece of software with a wonderful community but it has been painfully slow in receiving updates in the quickly evolving AI tooling space.
+`cecli` (probably pronounced like "Cecily") is yet another cli agent crafted for extensibility and customization. Originally a fork of the [Aider](https://cecli.dev/) AI pair programming tool, we aim to make agentic coding as maximally effective as it can be based on the growing capabilities of large language models.
 
-We aim to foster an open, collaborative ecosystem where new features, experiments, and improvements can be developed and shared rapidly. We believe in genuine FOSS principles and actively welcome contributors of all skill levels.
+We aim to foster an open, collaborative ecosystem where new features, experiments, and improvements can be developed and shared rapidly. We believe in the principles of FOSS and actively welcome contributors of all skill levels.
 
 If you are looking for bleeding-edge features or want to get your hands dirty with the internals of an AI coding agent, here's your sign.
+
 LLMs are a part of our lives from here on out so join us in learning about and crafting the future.
 
 ### Links

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/README.md

@@ -1,11 +1,11 @@
 ## Why `cecli`?
 
-`cecli` (probably pronounced like "Cecily", aka `aider-ce`) is a community-driven fork of the [Aider](https://cecli.dev/) AI pair programming tool.
-Aider is a fantastic piece of software with a wonderful community but it has been painfully slow in receiving updates in the quickly evolving AI tooling space.
+`cecli` (probably pronounced like "Cecily") is yet another cli agent crafted for extensibility and customization. Originally a fork of the [Aider](https://cecli.dev/) AI pair programming tool, we aim to make agentic coding as maximally effective as it can be based on the growing capabilities of large language models.
 
-We aim to foster an open, collaborative ecosystem where new features, experiments, and improvements can be developed and shared rapidly. We believe in genuine FOSS principles and actively welcome contributors of all skill levels.
+We aim to foster an open, collaborative ecosystem where new features, experiments, and improvements can be developed and shared rapidly. We believe in the principles of FOSS and actively welcome contributors of all skill levels.
 
 If you are looking for bleeding-edge features or want to get your hands dirty with the internals of an AI coding agent, here's your sign.
+
 LLMs are a part of our lives from here on out so join us in learning about and crafting the future.
 
 ### Links

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/benchmark/benchmark.py

@@ -1155,6 +1155,7 @@ async def run_test_real(
             use_enhanced_map=True,
             verbose=verbose,
             yes_always_commands=True,
+            max_reflections=0,
         ),
         map_mul_no_files=4,
         mcp_manager=None,

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/__init__.py

@@ -1,6 +1,6 @@
 from packaging import version
 
-__version__ = "0.97.1.dev"
+__version__ = "0.97.4.dev"
 safe_version = __version__
 
 try:

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.97.2'
-__version_tuple__ = version_tuple = (0, 97, 2)
+__version__ = version = '0.97.4'
+__version_tuple__ = version_tuple = (0, 97, 4)
 
-__commit_id__ = commit_id = 'g49fc44517'
+__commit_id__ = commit_id = 'g3f41a1bc9'

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/args.py

@@ -254,6 +254,18 @@ def get_parser(default_config_files, git_root):
             " If unspecified, defaults to the model's max_chat_history_tokens."
         ),
     )
+    group.add_argument(
+        "--max-reflections",
+        type=int,
+        default=3,
+        help="Maximum number of retries a model gets on malformed outputs (default: 3)",
+    )
+    group.add_argument(
+        "--cost-limit",
+        type=float,
+        default=None,
+        help="Cost limit per session, exceeding this forces prompt confirmation (default: None)",
+    )
     group.add_argument(
         "--file-diffs",
         action=argparse.BooleanOptionalAction,

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/coders/agent_coder.py

@@ -7,7 +7,7 @@ import platform
 import random
 import time
 import traceback
-from collections import Counter, defaultdict
+from collections import defaultdict
 from datetime import datetime
 from pathlib import Path
 
@@ -85,13 +85,18 @@ class AgentCoder(Coder):
         self.allowed_context_blocks = set()
         self.context_block_tokens = {}
         self.context_blocks_cache = {}
+        self.hot_reload_enabled = False
         self.tokens_calculated = False
         self.skip_cli_confirmations = False
         self.agent_finished = False
         self.agent_config = self._get_agent_config()
+        self._setup_agent()
         ToolRegistry.build_registry(agent_config=self.agent_config)
         super().__init__(*args, **kwargs)
 
+    def _setup_agent(self):
+        os.makedirs(".cecli/workspace", exist_ok=True)
+
     def _get_agent_config(self):
         """
         Parse and return agent configuration from args.agent_config.
@@ -119,6 +124,7 @@ class AgentCoder(Coder):
             config, "skip_cli_confirmations", nested.getter(config, "yolo", [])
         )
         config["command_timeout"] = nested.getter(config, "command_timeout", 30)
+        config["hot_reload"] = nested.getter(config, "hot_reload", False)
 
         config["tools_paths"] = nested.getter(config, "tools_paths", [])
         config["tools_includelist"] = nested.getter(
@@ -147,7 +153,7 @@ class AgentCoder(Coder):
 
         self.large_file_token_threshold = config["large_file_token_threshold"]
         self.skip_cli_confirmations = config["skip_cli_confirmations"]
-
+        self.hot_reload_enabled = config["hot_reload"]
         self.allowed_context_blocks = config["include_context_blocks"]
 
         for context_block in config["exclude_context_blocks"]:
@@ -259,6 +265,7 @@ class AgentCoder(Coder):
                         try:
                             parsed_args_list.append(json.loads(chunk))
                         except json.JSONDecodeError as e:
+                            self.model_kwargs = {}
                             self.io.tool_warning(
                                 f"Could not parse JSON chunk for tool {tool_name}: {chunk}"
                             )
@@ -320,6 +327,7 @@ class AgentCoder(Coder):
 
                 result_message = "\n\n".join(all_results_content)
             except Exception as e:
+                self.model_kwargs = {}
                 result_message = f"Error executing {tool_name}: {e}"
                 self.io.tool_error(f"""Error during {tool_name} execution: {e}
 {traceback.format_exc()}""")
@@ -843,6 +851,10 @@ I will proceed based on the tool results and updated context.""")
         self.files_edited_by_tools = set()
         return False
 
+    async def hot_reload(self):
+        if self.hot_reload_enabled:
+            self.skills_manager.hot_reload()
+
     async def _execute_tool_with_registry(self, norm_tool_name, params):
         """
         Execute a tool using the tool registry.
@@ -879,12 +891,9 @@ I will proceed based on the tool results and updated context.""")
         """
         Identifies repetitive tool usage patterns from rounds of tool calls.
 
-        This method combines count-based and similarity-based detection:
+        This method uses similarity-based detection:
         1. If the last round contained a write tool, it assumes progress and returns no repetitive tools.
-        2. It checks for any read tool that has been used 2 or more times across rounds.
-        3. If no tools are repeated, but all tools in the history are read tools,
-           it flags all of them as potentially repetitive.
-        4. It checks for similarity-based repetition using cosine similarity on tool call strings.
+        2. It checks for similarity-based repetition using cosine similarity on tool call strings.
 
         It avoids flagging repetition if a "write" tool was used recently,
         as that suggests progress is being made.
@@ -893,9 +902,6 @@ I will proceed based on the tool results and updated context.""")
         if history_len < 5:
             return set()
         similarity_repetitive_tools = self._get_repetitive_tools_by_similarity()
-        all_tools = []
-        for round_tools in self.tool_usage_history:
-            all_tools.extend(round_tools)
         if self.last_round_tools:
             last_round_has_write = any(
                 tool.lower() in self.write_tools for tool in self.last_round_tools
@@ -909,24 +915,14 @@ I will proceed based on the tool results and updated context.""")
                     if tool.lower() in self.read_tools or tool.lower() in self.write_tools
                 }
                 return filtered_similarity_tools if len(filtered_similarity_tools) else set()
-        if all(tool.lower() in self.read_tools for tool in all_tools):
-            # Only return tools that are in read_tools
-            return {tool for tool in all_tools if tool.lower() in self.read_tools}
-        tool_counts = Counter(all_tools)
-        count_repetitive_tools = {
-            tool
-            for tool, count in tool_counts.items()
-            if count >= 5 and tool.lower() in self.read_tools
-        }
         # Filter similarity_repetitive_tools to only include tools in read_tools or write_tools
         filtered_similarity_tools = {
             tool
             for tool in similarity_repetitive_tools
             if tool.lower() in self.read_tools or tool.lower() in self.write_tools
         }
-        repetitive_tools = count_repetitive_tools.union(filtered_similarity_tools)
-        if repetitive_tools:
-            return repetitive_tools
+        if filtered_similarity_tools:
+            return filtered_similarity_tools
         return set()
 
     def _get_repetitive_tools_by_similarity(self):
@@ -983,6 +979,27 @@ I will proceed based on the tool results and updated context.""")
 
         context_parts.append("\n\n")
         if repetitive_tools:
+            if not self.model_kwargs:
+                self.model_kwargs = {
+                    "temperature": (self.main_model.use_temperature or 1) + 0.1,
+                    "frequency_penalty": 0.2,
+                    "presence_penalty": 0.1,
+                }
+            else:
+                temperature = nested.getter(self.model_kwargs, "temperature")
+                freq_penalty = nested.getter(self.model_kwargs, "frequency_penalty")
+                if temperature and freq_penalty:
+                    self.model_kwargs["temperature"] = min(temperature + 0.1, 2)
+                    self.model_kwargs["frequency_penalty"] = min(freq_penalty + 0.1, 1)
+
+                if random.random() < 0.25:
+                    self.model_kwargs["temperature"] = max(temperature - 0.2, 1)
+                    self.model_kwargs["frequency_penalty"] = max(freq_penalty - 0.2, 0)
+
+            # One tenth of the time, just straight reset the randomness
+            if random.random() < 0.1:
+                self.model_kwargs = {}
+
             if self.turn_count - self._last_repetitive_warning_turn > 2:
                 self._last_repetitive_warning_turn = self.turn_count
                 self._last_repetitive_warning_severity += 1
@@ -1040,7 +1057,7 @@ I have detected repetitive usage of the following tools: {', '.join([f'`{t}`' fo
 
                 repetition_warning += f"""
 ### CRITICAL: Execution Loop Detected
-You are currently "spinning." To break the logic trap, you must:
+You are currently "spinning gears". To break the exploration loop, you must:
 1. **Analyze**: Use the `Thinking` tool to summarize exactly what you have found so far and why you were stuck.
 2. **Pivot**: Abandon or modify your current exploration strategy. Try focusing on different files or running tests.
 3. **Reframe**: To ensure your logic reset, include a 2-sentence story about {animal} {verb} {fruit} in your thoughts.
@@ -1049,6 +1066,9 @@ Prioritize editing or verification over further exploration.
                 """
 
             context_parts.append(repetition_warning)
+        else:
+            self.model_kwargs = {}
+
         context_parts.append("</context>")
         return "\n".join(context_parts)
 

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/coders/architect_coder.py

@@ -1,7 +1,7 @@
 import asyncio
 
 from ..commands import SwitchCoderSignal
-from ..helpers.conversation import ConversationManager, MessageTag
+from ..helpers.conversation import ConversationManager
 from .ask_coder import AskCoder
 from .base_coder import Coder
 
@@ -61,9 +61,7 @@ class ArchitectCoder(AskCoder):
         editor_coder = await Coder.create(**new_kwargs)
 
         # Re-initialize ConversationManager with editor coder
-        ConversationManager.initialize(
-            editor_coder, reset=True, reformat=True, preserve_tags=[MessageTag.DONE, MessageTag.CUR]
-        )
+        ConversationManager.initialize(editor_coder, reset=True, reformat=True, preserve_tags=True)
 
         if self.verbose:
             editor_coder.show_announcements()
@@ -84,7 +82,7 @@ class ArchitectCoder(AskCoder):
                 original_coder or self,
                 reset=True,
                 reformat=True,
-                preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                preserve_tags=True,
             )
 
             self.total_cost = editor_coder.total_cost
@@ -96,7 +94,7 @@ class ArchitectCoder(AskCoder):
                 original_coder or self,
                 reset=True,
                 reformat=True,
-                preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                preserve_tags=True,
             )
 
         raise SwitchCoderSignal(main_model=self.main_model, edit_format="architect")

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/coders/base_coder.py

@@ -32,7 +32,7 @@ from uuid import uuid4 as generate_unique_id
 
 import httpx
 from litellm import experimental_mcp_client
-from litellm.types.utils import ChatCompletionMessageToolCall, Function, ModelResponse
+from litellm.types.utils import ModelResponse
 from prompt_toolkit.patch_stdout import patch_stdout
 from rich.console import Console
 
@@ -40,7 +40,7 @@ import cecli.prompts.utils.system as prompts
 from cecli import __version__, models, urls, utils
 from cecli.commands import Commands, SwitchCoderSignal
 from cecli.exceptions import LiteLLMExceptions
-from cecli.helpers import command_parser, coroutines, nested
+from cecli.helpers import command_parser, coroutines, nested, responses
 from cecli.helpers.conversation import (
     ConversationChunks,
     ConversationManager,
@@ -156,6 +156,8 @@ class Coder:
     tool_reflection = False
     last_user_message = ""
     uuid = ""
+    model_kwargs = {}
+    cost_multiplier = 1
 
     # Task coordination state variables
     input_running = False
@@ -210,34 +212,14 @@ class Coder:
 
             use_kwargs = dict(from_coder.original_kwargs)  # copy orig kwargs
 
-            # If the edit format changes, we can't leave old ASSISTANT
-            # messages in the chat history. The old edit format will
-            # confused the new LLM. It may try and imitate it, disobeying
-            # the system prompt.
-            # Get DONE messages from ConversationManager
-            done_messages = ConversationManager.get_messages_dict(MessageTag.DONE)
-            if edit_format != from_coder.edit_format and done_messages and summarize_from_coder:
-                try:
-                    io.tool_warning("Summarizing messages, please wait...")
-                    done_messages = await from_coder.summarizer.summarize_all(done_messages)
-                except (KeyboardInterrupt, ValueError):
-                    # If summarization fails, keep the original messages and warn the user
-                    io.tool_warning(
-                        "Chat history summarization failed, continuing with full history"
-                    )
-
-            # Bring along context from the old Coder
-            # Get CUR messages from ConversationManager
-            cur_messages = ConversationManager.get_messages_dict(MessageTag.CUR)
-
             update = dict(
                 fnames=list(from_coder.abs_fnames),
                 read_only_fnames=list(from_coder.abs_read_only_fnames),  # Copy read-only files
                 read_only_stubs_fnames=list(
                     from_coder.abs_read_only_stubs_fnames
                 ),  # Copy read-only stubs
-                done_messages=done_messages,
-                cur_messages=cur_messages,
+                done_messages=[],
+                cur_messages=[],
                 coder_commit_hashes=from_coder.coder_commit_hashes,
                 commands=from_coder.commands.clone(),
                 total_cost=from_coder.total_cost,
@@ -380,6 +362,9 @@ class Coder:
 
         self.context_compaction_max_tokens = context_compaction_max_tokens
         self.context_compaction_summary_tokens = context_compaction_summary_tokens
+        self.max_reflections = (
+            3 if self.edit_format == "agent" else nested.getter(self.args, "max_reflections", 3)
+        )
 
         if not fnames:
             fnames = []
@@ -411,22 +396,6 @@ class Coder:
         self.add_gitignore_files = add_gitignore_files
         self.abs_read_only_stubs_fnames = set()
 
-        # Always use ConversationManager as the source of truth
-        # Add any provided messages to ConversationManager
-        if done_messages:
-            for msg in done_messages:
-                ConversationManager.add_message(
-                    message_dict=msg,
-                    tag=MessageTag.DONE,
-                )
-
-        if cur_messages:
-            for msg in cur_messages:
-                ConversationManager.add_message(
-                    message_dict=msg,
-                    tag=MessageTag.CUR,
-                )
-
         self.io = io
         self.io.coder = weakref.ref(self)
 
@@ -1612,9 +1581,23 @@ class Coder:
             self.reflected_message = None
             self.tool_reflection = False
 
+            if float(self.total_cost) > self.cost_multiplier * (
+                nested.getter(self.args, "cost_limit", float("inf")) or float("inf")
+            ):
+                if await self.io.confirm_ask(
+                    "You have reached your configured cost limit. Continue?",
+                    group_response="Cost Limit",
+                    explicit_yes_required=True,
+                ):
+                    Coder.cost_multiplier += 1
+                else:
+                    return
+
             async for _ in self.send_message(message):
                 pass
 
+            await self.hot_reload()
+
             if not self.reflected_message:
                 await self.auto_save_session(force=True)
                 break
@@ -2399,7 +2382,7 @@ class Coder:
                 force=True,  # Force update existing message
             )
 
-        if edited and self.auto_test:
+        if edited and self.auto_test and self.test_cmd:
             test_errors = await self.commands.execute("test", self.test_cmd)
             self.test_outcome = not test_errors
             if test_errors:
@@ -2750,6 +2733,9 @@ class Coder:
     async def reply_completed(self):
         pass
 
+    async def hot_reload(self):
+        pass
+
     async def show_exhausted_error(self):
         output_tokens = 0
         if self.partial_response_content:
@@ -2995,6 +2981,7 @@ class Coder:
                 self.temperature,
                 # This could include any tools, but for now it is just MCP tools
                 tools=tools,
+                override_kwargs=self.model_kwargs,
             )
             self.chat_completion_call_hashes.append(hash_object.hexdigest())
 
@@ -3166,7 +3153,7 @@ class Coder:
                         reasoning_content = None
 
                 if reasoning_content:
-                    if nested.getter(self, "args.show_thinking"):
+                    if nested.getter(self.args, "show_thinking"):
                         if not self.got_reasoning_content:
                             text += f"<{REASONING_TAG}>\n\n"
                         text += reasoning_content
@@ -3202,7 +3189,7 @@ class Coder:
                 self.stream_wrapper(content_to_show, final=False)
             elif text:
                 # Apply reasoning tag formatting for non-pretty output
-                if nested.getter(self, "args.show_thinking"):
+                if nested.getter(self.args, "show_thinking"):
                     text = replace_reasoning_tags(text, self.reasoning_tag_name)
                 try:
                     self.stream_wrapper(text, final=False)
@@ -3324,66 +3311,16 @@ class Coder:
         # If no native tool calls, check if the content contains JSON tool calls
         # This handles models that write JSON in text instead of using native calling
         if not self.partial_response_tool_calls and self.partial_response_content:
-            try:
-                # Simple extraction of JSON-like structures that look like tool calls
-                # Only look for tool calls if it looks like JSON
-                if "{" in self.partial_response_content or "[" in self.partial_response_content:
-                    json_chunks = utils.split_concatenated_json(self.partial_response_content)
-                    extracted_calls = []
-                    chunk_index = 0
-
-                    for chunk in json_chunks:
-                        chunk_index += 1
-                        try:
-                            json_obj = json.loads(chunk)
-                            if (
-                                isinstance(json_obj, dict)
-                                and "name" in json_obj
-                                and "arguments" in json_obj
-                            ):
-                                # Create a Pydantic model for the tool call
-                                function_obj = Function(
-                                    name=json_obj["name"],
-                                    arguments=(
-                                        json.dumps(json_obj["arguments"])
-                                        if isinstance(json_obj["arguments"], (dict, list))
-                                        else str(json_obj["arguments"])
-                                    ),
-                                )
-                                tool_call_obj = ChatCompletionMessageToolCall(
-                                    type="function",
-                                    function=function_obj,
-                                    id=f"call_{len(extracted_calls)}_{int(time.time())}_{chunk_index}",
-                                )
-                                extracted_calls.append(tool_call_obj)
-                            elif isinstance(json_obj, list):
-                                for item in json_obj:
-                                    if (
-                                        isinstance(item, dict)
-                                        and "name" in item
-                                        and "arguments" in item
-                                    ):
-                                        function_obj = Function(
-                                            name=item["name"],
-                                            arguments=(
-                                                json.dumps(item["arguments"])
-                                                if isinstance(item["arguments"], (dict, list))
-                                                else str(item["arguments"])
-                                            ),
-                                        )
-                                        tool_call_obj = ChatCompletionMessageToolCall(
-                                            type="function",
-                                            function=function_obj,
-                                            id=f"call_{len(extracted_calls)}_{int(time.time())}_{chunk_index}",
-                                        )
-                                        extracted_calls.append(tool_call_obj)
-                        except json.JSONDecodeError:
-                            continue
-
-                    if extracted_calls:
-                        self.partial_response_tool_calls = extracted_calls
-            except Exception:
-                pass
+            extracted_calls = responses.extract_tools_from_content_json(
+                self.partial_response_content
+            )
+            if not extracted_calls:
+                extracted_calls = responses.extract_tools_from_content_xml(
+                    self.partial_response_content
+                )
+
+            if extracted_calls:
+                self.partial_response_tool_calls = extracted_calls
 
         return response, func_err, content_err
 
@@ -3405,7 +3342,7 @@ class Coder:
     def live_incremental_response(self, final):
         show_resp = self.render_incremental_response(final)
         # Apply any reasoning tag formatting
-        if nested.getter(self, "args.show_thinking"):
+        if nested.getter(self.args, "show_thinking"):
             show_resp = replace_reasoning_tags(show_resp, self.reasoning_tag_name)
 
         # Track streaming state to avoid repetitive output

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/commands/agent_model.py

@@ -3,7 +3,7 @@ from typing import List
 import cecli.models as models
 from cecli.commands.utils.base_command import BaseCommand
 from cecli.commands.utils.helpers import format_command_result
-from cecli.helpers.conversation import ConversationManager, MessageTag
+from cecli.helpers.conversation import ConversationManager
 
 
 class AgentModelCommand(BaseCommand):
@@ -69,7 +69,7 @@ class AgentModelCommand(BaseCommand):
                 temp_coder,
                 reset=True,
                 reformat=True,
-                preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                preserve_tags=True,
             )
 
             verbose = kwargs.get("verbose", False)
@@ -86,7 +86,7 @@ class AgentModelCommand(BaseCommand):
                     original_coder,
                     reset=True,
                     reformat=True,
-                    preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                    preserve_tags=True,
                 )
 
                 # Restore the original model configuration

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/commands/editor_model.py

@@ -3,7 +3,7 @@ from typing import List
 import cecli.models as models
 from cecli.commands.utils.base_command import BaseCommand
 from cecli.commands.utils.helpers import format_command_result
-from cecli.helpers.conversation import ConversationManager, MessageTag
+from cecli.helpers.conversation import ConversationManager
 
 
 class EditorModelCommand(BaseCommand):
@@ -69,7 +69,7 @@ class EditorModelCommand(BaseCommand):
                 temp_coder,
                 reset=True,
                 reformat=True,
-                preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                preserve_tags=True,
             )
 
             verbose = kwargs.get("verbose", False)
@@ -86,7 +86,7 @@ class EditorModelCommand(BaseCommand):
                     original_coder,
                     reset=True,
                     reformat=True,
-                    preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                    preserve_tags=True,
                 )
 
                 # Restore the original model configuration

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/commands/model.py

@@ -3,7 +3,7 @@ from typing import List
 import cecli.models as models
 from cecli.commands.utils.base_command import BaseCommand
 from cecli.commands.utils.helpers import format_command_result
-from cecli.helpers.conversation import ConversationManager, MessageTag
+from cecli.helpers.conversation import ConversationManager
 
 
 class ModelCommand(BaseCommand):
@@ -73,7 +73,7 @@ class ModelCommand(BaseCommand):
                 temp_coder,
                 reset=True,
                 reformat=True,
-                preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                preserve_tags=True,
             )
 
             verbose = kwargs.get("verbose", False)
@@ -90,7 +90,7 @@ class ModelCommand(BaseCommand):
                     original_coder,
                     reset=True,
                     reformat=True,
-                    preserve_tags=[MessageTag.DONE, MessageTag.CUR],
+                    preserve_tags=True,
                 )
 
                 # Restore the original model configuration

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/commands/save_session.py

@@ -25,9 +25,12 @@ class SaveSessionCommand(BaseCommand):
     @classmethod
     def get_completions(cls, io, coder, args) -> List[str]:
         """Get completion options for save-session command."""
-        # For save-session, we could return existing session names for completion
-        # For now, return empty list
-        return []
+        # Return existing session names for completion to prevent accidental overwrites
+        from cecli import sessions
+
+        session_manager = sessions.SessionManager(coder, io)
+        sessions_list = session_manager.list_sessions()
+        return [session_info["name"] for session_info in sessions_list]
 
     @classmethod
     def get_help(cls) -> str:
@@ -40,4 +43,8 @@ class SaveSessionCommand(BaseCommand):
         help_text += "  /save-session bug-fix         # Save session as 'bug-fix'\n"
         help_text += "\nSessions are saved in the .cecli/sessions/ directory as JSON files.\n"
         help_text += "Use /list-sessions to see saved sessions and /load-session to load them.\n"
+        help_text += (
+            "\nNote: Existing session names will be shown for tab completion to help prevent"
+            " accidental overwrites.\n"
+        )
         return help_text

{cecli_dev-0.97.2 → cecli_dev-0.97.4}/cecli/commands/tokens.py

@@ -64,6 +64,7 @@ class TokensCommand(BaseCommand):
         tokens_done = 0
         tokens_cur = 0
         tokens_diffs = 0
+        tokens_file_contexts = 0
 
         if msgs_done:
             tokens_done = coder.main_model.token_count(msgs_done)